Soil organic carbon stocks in the Limpopo National Park, Mozambique: Amount, spatial distribution and uncertainty

Abstract

Many areas in sub-Saharan African are data-poor and poorly accessible. The estimation of soil organic carbon (SOC) stocks in these areas will have to rely on the limited available secondary data coupled with restricted field sampling. We assessed the total SOC stock, its spatial variation and the causes of this variation in Limpopo National Park (LNP), a data-poor and poorly accessible area in southwestern Mozambique. During a field survey, A-horizon thickness was measured and soil samples were taken for the determination of SOC concentrations. SOC concentrations were multiplied by soil bulk density and A-horizon thickness to estimate SOC stocks. Spatial distribution was assessed through: i) a measure-and-multiply approach to assess average SOC stocks by landscape unit, and ii) a soil-landscape model that used soil forming factors to interpolate SOC stocks from observations to a grid covering the area by ordinary (OK) and universal (UK) kriging. Predictions were validated by both independent and leave-one-out cross validations. The total SOC stock of the LNP was obtained by i) calculating an area-weighted average from the means of the landscape units and by ii) summing the cells of the interpolated grid. Uncertainty was evaluated by the mean standard error for the measure-and-multiply approach and by the mean kriging prediction standard deviation for the soil-landscape model approach. The reliability of the estimates of total stocks was assessed by the uncertainty of the input data and its effect on estimates. The mean SOC stock from all sample points is 1.59kgm−2; landscape unit averages are 1.13–2.46kgm−2. Covariables explained 45% (soil) and 17% (coordinates) of SOC stock variation. Predictions from spatial models averaged 1.65kgm−2 and are within the ranges reported for similar soils in southern Africa. The validation root mean square error of prediction (RMSEP) was about 30% of the mean predictions for both OK and UK. Uncertainty is high (coefficient of variation of about 40%) due to short-range spatial structure combined with sparse sampling. The range of total SOC stock of the 10,410km−2 study area was estimated at 15,579–17,908Gg. However, 90% confidence limits of the total stocks estimated are narrower (5–15%) for the measure-and-multiply model and wider (66–70%) for the soil-landscape model. The spatial distribution is rather homogenous, suggesting levels are mainly determined by regional climate.